nandan

Physics → software → quantum. Interested in how information becomes knowledge. Exploring at the intersection of quantum computing and machine intelligence.

[Ilir Aliu @IlirAliu_]

ETH Zurich just open-sourced their entire 2026 robot learning course.

Not a MOOC. The actual course. Slides, lecture recordings, coding assignments, GitHub repo.

The curriculum goes from imitation learning and RL all the way to Vision-Language-Action models and foundation models for robotics.

Guest lectures from the co-founder of Physical Intelligence. The creator of Diffusion Policy. Pieter Abbeel. Dieter Fox.

12 weeks. Free. No signup.

If you want to understand where robot intelligence is actually heading… this is the reading list the field is using right now.

📍[https://t.co/eKsIjILi60]

——

Weekly robotics and AI insights.
Subscribe free: https://t.co/9Nm01QUcw3

[nandan @nandanito]

https://t.co/0XF0QeZJy0

50 curated circuits to fork and run. Or push your own. I'm reviewing early-beta feedback closely — Submit Feedback is the best place to leave it.

The next month is about what real users surface.

[nandan @nandanito]

The hardest part wasn't the quantum — it was making the developer experience feel like the tools we already love.

git push over HTTPS for a circuit repo. Cirq + PennyLane runners that match Qiskit's UX. Stuff that should be table stakes but rarely is.

[nandan @nandanito]

I work in distributed systems and digital-asset infrastructure. Quantum has been the parallel thread for a while — independent research, and tooling I wanted while learning didn't exist.

If you've been quantum-curious, here's something to poke at without credentials or a PhD:

[nandan @nandanito]

What works today:

• git clone / git push circuits over HTTPS
• Native Qiskit, Cirq, PennyLane execution
• Fork + star + share results
• Run pages that unfurl on socials with previews

It's a beta. It's also genuinely usable.

[nandan @nandanito]

Six months ago I started learning quantum computing.

What I needed didn't exist — a place where I could push a circuit, fork someone else's, run it, share the result. GitHub for quantum.

So I built it. Today QubitHub beta opens to first builders.

https://t.co/0XF0QeZJy0

[nandan @nandanito]

A few months ago I watched a coding agent delete a file I cared about, recreate a stripped-down version three steps later, and report success.

The tests passed because the deletion had taken the test for that file with it.

I rolled back, but the question that stuck with me was not “what did it do?” I had tool traces for that.

The question was:

What did the agent think it was doing?

I had OTel traces. Prompt logs. Conversation transcripts. Memory snapshots.

More telemetry than I could read.

But I had no record of belief.

That is the gap I keep seeing in agent systems.

Runtime tells us what the agent did.
Memory tells us what it retained.
Observability tells us what we can inspect after the fact.

But none of these cleanly records what the agent took to be true when it acted.

Sometimes the tool call is valid. The retrieved context is present. The trace is clean. But somewhere, a claim became an actionable belief — and that promotion was never recorded.

I wrote the first post in a short series on this problem:

“The question my coding agent couldn’t answer”

https://t.co/u8BE0JSGYf

Part 2 will cover the architecture I’m building to make that belief/action chain explicit.

#AIAgents #DeveloperTools #AISafety

[Alibaba Cloud @alibaba_cloud]

Introducing Qwen3.6-Max-Preview, early preview of our next flagship model

·Improved agentic coding capability over Qwen3.6-Plus
·Stronger world knowledge and instruction following
·Improved real-world agent and knowledge reliability performance

Smarter, sharper, still evolving.

[Anastasia Marchenkova @amarchenkova]

When Ivan Deutsch does a podcast, I highly recommend listening. He's one of the field's most eminent quantum information theorists, and he's challenging the field's consensus on two-level quantum computing systems.

We haven't finished with qubits, but we are already talking about qudits!

[Anastasia Marchenkova @amarchenkova]

Europe is diving into quantum manufacturing infrastructure!

The €50M SUPREME project officially launched last week, uniting 23 partners to scale superconducting quantum technologies, including advanced fabrication processes, with academic, industry, and research partners including VTT, Delft, University of Naples Federico II, Infineon Technologies, IQM, Alice & Bob, QuantWare, QphoX, Peak Quantum, and many more.

https://t.co/2gZx5w2ims

[Anastasia Marchenkova @amarchenkova]

This paper constructs a universal shortest analytic quantum algorithm for arbitrary diagonal matrices of any size, derived via interpretable ML. Crucial for quantum circuit optimization.

https://t.co/BIdf9skZ16

[Anastasia Marchenkova @amarchenkova]

New work from ParityQC and University of Innsbruck introduces a fault-tolerant quantum computing architecture that significantly reduces resource overhead for universal quantum computation.

The T gate is the standard non-Clifford gate underlying universal quantum computing — and the most expensive thing to make fault-tolerant, requiring dedicated "magic state distillation" factories that dominate projected qubit budgets. Algorithms like Shor's, the quantum Fourier transform, and phase estimation actually need much smaller-angle rotations than T, conventionally built by stringing together long sequences of T gates as approximations. This work distills those smaller rotations directly, skipping the approximation step.

⚛️ targets distillation of non-Clifford gates, critical for universal fault tolerance.
⚛️ extends beyond the standard T gate to distill rotation gates at deeper levels of the Clifford hierarchy.
⚛️ for algorithms like the quantum Fourier transform and phase estimation, reduces overhead for small-angle rotations down to T^(1/32).
⚛️ when distilling T together with √T for arbitrary small-angle rotations, cuts resource overhead by 26% and minimum logical error rate by 43%, versus parity-unfolded T-only distillation.
⚛️ compatible with 2D planar qubit architectures, enabling fault-tolerant operations with nearest-neighbor interactions.
⚛️ optimized for noise-biased platforms.

This tackles a core challenge in building scalable fault-tolerant quantum computers, cutting resource overhead and improving gate fidelity on planar, noise-biased hardware.

details in a preprint, pending peer review.

[Aakash Gupta @aakashgupta]

Yann LeCun closed $1.03B for AMI Labs on March 10. Three days later, this paper dropped from his NYU collaborators.

15M parameters. Single GPU. A few hours of training.

LeWorldModel is the first JEPA that trains end-to-end from raw pixels. Two loss terms: predict the next embedding, keep the latent space Gaussian. Previous JEPAs needed exponential moving averages or pretrained encoders to avoid representation collapse. LeWM doesn't.

Six hyperparameters down to one.

The numbers are the story. Foundation-model-based world models require hundreds of millions of parameters and serious compute to plan a control task. LeWM plans up to 48x faster while staying competitive on 2D and 3D benchmarks. The whole thing fits on a laptop GPU.

Look at the trajectory. Yann announced his Meta departure in November 2025 after 12 years and called founding FAIR his "proudest non-technical accomplishment." On March 10, 2026, AMI Labs closed the largest seed round in European history at a $3.5B pre-money valuation. Bezos, Nvidia, Samsung, and Toyota all wrote checks.

Three days later: a paper showing that JEPA-from-pixels is no longer fragile and no longer compute-heavy. The engineering scaffolding that made it look like an academic curiosity is gone.

The authors sit at Mila, NYU, Samsung SAIL, and Brown. None at Meta.

Yann's bet was that the path to machine intelligence runs through world models, not language models. He left a public company to build it. Each JEPA paper from his network resets the assumed cost structure for that bet. This one makes world modeling laptop-cheap.

Meta still has the GPUs. The architecture left.

[nandan @nandanito]

Nice to be part of it.

DeFi needs a robust Privacy Layer. @zksdk_labs is the right way to do it.

[Algebra DEX Engine @CryptoAlgebra]

Algebra is announcing a new integration: @zksdk_labs's Privacy Layer is live on testnet! 🔨

Enable private swaps & transfers, fully isolated environment, no risk to the core AMM.

Launch your own anonymity set or plug into a shared pool for:

• custom anonymity sets
• shared privacy pools
• widget / API / SDK integrations

More details coming soon!

[zksdk @zksdk_labs]

Introducing zkSDK. Privacy for DeFi.

No new chain.
No migration.
No rebuild.

zkSDK plugs into existing infrastructure.

Coming to Algebra. More Soon.

[Google Quantum AI @GoogleQuantumAI]

Our team ran a verifiable quantum algorithm that probes how parts of a quantum system interact, from molecules to magnets and beyond. On our Willow chip, it ran 13,000× faster than the best classical supercomputers. A first in quantum computing → https://t.co/j56g2M7gx0

[Massimo @Rainmaker1973]

Scientists have created one of the most detailed 3D reconstructions of a human cell (eukaryotic cell) ever produced.

This groundbreaking model, often termed a "Cellular Landscape Cross-Section Through a Eukaryotic Cell," combines data from X-ray tomography, nuclear magnetic resonance (NMR), and cryo-electron microscopy to map molecular structures in extreme detail.

[Anastasia Marchenkova @amarchenkova]

Google is opening Willow, their 105-qubit quantum processor, to outside researchers through a curated Early Access Program.

Proposals due May 15, 2026, selections July 1.

Willow is the processor behind Google's quantum echoes — 99.97% single-qubit fidelity, 99.88% 2-qubit, which ran 13,000x faster than the best classical simulation.

[Steve Weis @sweis]

New Quantum Computing Landscape for 2026 from @sejaques: https://t.co/9BNEtbP32N

[nandan @nandanito]

I've been learning quantum computing on the side for a few months. The thing that kept tripping me up wasn't the math — it was that a surprising fraction of "canonical" circuits from papers and tutorials would fail quietly. Silently broken Hamiltonians. Reversed bit orders. Post-deprecation API drift. You run the code, get a plausible-looking number, and have no way to tell if it was right.

So I started building a personal library where every circuit is checked against its reference output. That's now open to browse — https://t.co/DyTICbGVAB . No signup required: 50+ curated circuits across Qiskit, PennyLane, and Cirq, each one audited against its reference output before it's listed.

A few things I'm happy about:

- Readable URLs — every circuit lives at `https://t.co/wbwPuugTYn{owner}/{name}`, no UUIDs
- Audited — every circuit has been verified against its reference output (the number I had to fix is a separate post)
- Multi-framework — the same algorithm can show up in Qiskit, PennyLane, and Cirq side by side

Sign-up's open today — you get star and save. Creating your own circuit is still invite-only while I stabilize the builder flow; drop me a line after you sign up and I'll flip the switch as slots free up. In-browser execution is the next thing I'm finishing — separate post when it's live.

Six months in, this is the thing I wish had existed when I started. More soon.